SiFive has the first mover’s advantage in creating a business around producing chips based on the RISC-V open instruction set. Yet, its primary focus is its SaaS based on-cloud chip customisation service. Huzefa Cutlerywala, MD, Open-Silicon India (a SiFive company), explains to Rahul Chopra of the EFY Group how Open-Silicon and SiFive structure their business strategies and how their SaaS offering will benefit the design community in creating custom silicon.
EB: How does your three-layer business model shape your business – with one layer selling development boards, the second selling chips and the third offering services around the whole ecosystem?
The first layer that you mentioned is more of an enabler or a marketing expenditure and less of a revenue generation process. The big business is in the other two levels. The intention is to provide software, tools and applications that deliver the ability to make chips, on the cloud. This is how the company plans to grow.
EB: Is it similar to the Red Hat model?
To some extent, it is. Red Hat is selling its own version of Linux, which is analogous to our second business model, i.e., the chip model. Customers who are interested in making their own CPU are going to buy the core designer and design different cores. Some big companies (like SanDisk or Western Digital) are going to take the core and will make their chips themselves. We are looking out for smaller companies for whom Open-Silicon becomes the platform to take the ideas, make the complete chip and then ship it.
EB: Won’t other independent chip design houses that use your open source platform end up becoming competitors to your chip business?
Our primary business focus is our SaaS offering, i.e., delivering the ability to make chips on the cloud. Let’s say some companies are in the SSD business and want to make their own exclusive chips. Enabling that would be our primary business. We will love it if all the design houses in India come and start using our model. As more people start using our platform, more people will make better chips. The ability to make chips and create custom silicon is an enabler and a key value-add for our customers. Over the years, it can become our big business, while our customers will be happy making their own chips. In fact, if CEOs and CTOs are not going to look into this, they may start falling behind.
EB: Is training/certification part of your plans to generate additional revenue?
That may happen. But if you think about it from the revenue perspective, the big elements are going to be the product based models. The semiconductor industry has an annual turnover of about US$ 400 to US$ 500 billion. If we can build our business around getting just 10 per cent of the market to use our services to develop custom SoCs, it will be a big step ahead. Therefore, we are focused more towards the custom chips. We want our customers to make chips easily and that’s where our SaaS model cloud platform comes in. It’s a Saas model-plus-silicon based business.
EB: For a layman, is there any difference between SiFive and Open-Silicon or is it just in the brand names?
The core configuration and the cloud connectivity is done through SiFive infrastructure. However, we want to leverage the reputation that the Open-Silicon brand has built over the years to let people know that we build reliable chips. SiFive has more expertise in the CPU domain while Open-Silicon’s strengths are in chip creation. To illustrate this difference, let’s imagine an SoC factory and a chip factory. The expertise of the core factory (making a customised CPU core) comes from SiFive. The chip expertise lies with Open-Silicon. The SoC is developed based on variable models from both sides.
EB: How does SiFive deliver the core expertise on the cloud?
Users can log in through our SaaS model and play around with the different cores available digitally on the platform. These can be customised to one’s particular application. Depending on the series of our offerings being used (Series E2, E3, E5…), one can avail various options and tap into various Power Performance Area numbers. Once the optimal choice has been finalised, a mere click confirms the core.
Now that the core is made, it can be taken to the SoC factory where the interfaces, ADCs, the memory, USB connections, etc, are. The program behind this system is called Chizel, created by Berkeley professors and students. This program does not use the generic Verilog language but another language developed at Berkeley.
EB: How significant a role does India play in this overall scheme of engineering for Open-Silicon and SiFive?
When it started off in 2003 coming out of Intel Microelectronics, Open-Silicon had plans to design in India. Many layouts were done out of the Bengaluru office. Since then we have been working out of India. We have the physical design centre in Bengaluru where today we have about 150 engineers doing innovative work like 7nm-16nm designs. About 10 years ago we had set up another centre in Pune which now has about 50 engineers. That unit focuses on helping customers with the chip design element so that they do not have to be experts in everything.
EB: How important is India as a sales market?
We do not have any particular sales team in India as of now, but there are plans to change things. We have been in India long enough. People who want to do a chip design know they can go to a lot of design service players and get the design done. But on the manufacturing side, there are not many companies. Which is why, automatically, a lot of people decide to call us.
EB: Do you have any tie-up with distributors like Mouser or Digi-Key for these boards?
There are plans for this in the near future. Students, professors and startups are showing interest. In fact, at present we are running out of boards and I believe that is a good position to be in. So there is high demand and we are trying to keep up with it by exploring the possibilities of such tie-ups.
EB: Does the RISC-V Foundation have an India chapter or any office bearer here?
I do not know if there is a separate Indian chapter for the foundation. If an Indian company wants to become a member, it has to route it application through the US foundation. From India, IIT Madras is a member of the RISC-V Foundation and is involved with us. In July, one of the RISC-V workshops was held at IIT-M. C-DAC is also a member.
EB: Who were the early adopters of RISC-V in India, and wasn’t RISC-V declared as the national Instruction Set Architecture (ISA) in the latest open source announcements?
The initiative was very specific to RISC-V ISA, in fact. When we communicated with IIT-Madras professors, we understood that they were looking for something standalone that they could use in their own labs. When they came to know about an open ISA like RISC-V being put out there, they were the early birds to adopt it. They understood the value of the open instruction set, which is why they have their own architecture today. We know that the government has asked its departments to adopt RISC-V whenever they are building something of their own.
EB: How is the RISC-V community expanding?
With the model of evangelising we are going ahead with, we are seeing new universities, institutions and smaller companies joining regularly, given the economical nature of joining the committee. The more people start joining and leveraging the modular ISA, the more test suites will be added to evaluate any customised CPU that is built. RISC-V is just in its infancy and has a long way to go. The organisation has 180 members globally. There is a fee to join the foundation, which varies based on the size and number of members in a company.
EB: Will there be any handholding from Open-Silicon’s India office for startups or OEMs who want to do their own customisation?
It depends on what level of application they are looking for. I think, in the future, we’re definitely going to spend a lot more as an organisation to reach out to academia in India as well as overseas. That is a stated programme by our CEO who himself was a professor. He can very well see the importance of it. Students will be able to learn the SiFive RISC-V Core generator, and four to six years down the line, come out of universities with the necessary knowledge. Then there are people who have their own companies and who drive innovation. For all of them, SiFive has made certain tools available which can be downloaded and tried out. Later, when they realise that the software is good and there is a market for it, they can come back to us and we will help them in getting more done.
EB: Do you have your own production facility for ASICs?
We are a value chain aggregator. We have partnered with some of the biggest foundries for production, like with Taiwan Semiconductor Manufacturing Company (TSMC).
EB: Will your engineers help TSMC manufacture the chips?
Yes we will. It’s very hard when you get into the smaller dimensions like 45nm, 28nm, 16nm and 7nm, where a straight line is no more straight and starts becoming jagged since you reach atomic levels. Our team in Taiwan manages the production, starting from prototypes to thousands of wafers, selling millions of parts.
EB: Are there any specific application areas that will create higher demand for SiFive chipsets in particular?
From our experience, these are more used in high performance computing, servers, storage and other such high-tech use cases where custom silicon precisely fits in.
Of course, there is the promise of IoT. However, from the custom silicon perspective, it has not yet materialised. Many products have grown using existing microcontrollers and processors – but not custom silicon. Hopefully, more processors will become available with RISC-V and SiFive, whereby we can help make chips in a shorter time and in a less expensive manner for IoT applications as well.
EB: So has no IoT startup approached you till now for customised chips?
We received proposals from about seven to ten IoT application makers to build custom chips. But after talking to them, we realised that in not even one case did it make any business sense to explore a custom chip. For IoT applications, areas like smart metering, leakage sensors, surveillance, etc, are all niche applications with low volumes. You need to ensure the returns on the investment you make in these businesses before wondering whether you need a custom chip or not.
EB: Is custom silicon about reducing cost or protecting IP?
Both! In the IoT space, there are four clear reasons why they use custom chips. The first is cost. The second is the form factors and features that one wants. The third is about security capacities. And the final requirement is power efficiency. Moreover, the customers get their logo printed on the chip, which itself creates an IP.
